Found before it was designed

Take an empty drink can and press down on it. It does not crumple randomly. It collapses into a regular lattice of diamonds, the same lattice every time, and that lattice is a crease pattern which satisfies the flat-folding theorems at every vertex.

Nobody designed it. A buckling load chose it, on the grounds that it is the lowest-energy way for a thin cylindrical shell to fail, and Yoshimura published the geometry in 1951 in an aeronautics report about exactly that failure.

What the shell is doing #

The mechanics are worth a paragraph because they explain why the pattern is regular rather than arbitrary.

A thin cylindrical shell under axial compression has two ways to fail. It can buckle axisymmetrically, folding into rings like a concertina, and it can buckle into a diamond lattice. Which one happens depends on the geometry, and for thin shells the diamond mode is at a lower load.

The mode is a shape, and it is a solution of a minimisation: among the ways the shell can deform, the one that costs least energy. The diamonds are regular because the minimisation is translation-invariant around the cylinder and along it, so the answer repeats.

The pattern is flat-foldable, and that is surprising #

Here is the part that should not be taken for granted, because the site’s own machinery makes it checkable.

The shell was minimising elastic energy. Nothing in that calculation knows about Kawasaki’s condition, Maekawa’s condition, or the big-little-big lemma. There is no reason of principle why a buckling mode should be a flat-foldable crease pattern — plenty of deformations are not.

It is one anyway. Every interior vertex of the Yoshimura pattern satisfies all four local conditions, and the generator asserts it rather than asserting that it was checked.

Why it is flat-foldable #

The reason is not mysterious once stated and it is worth stating, because it explains a whole family of found patterns.

The shell is thin and does not stretch appreciably. A deformation that stretched the surface would cost far more energy than one that bends it, so the minimiser is very nearly an isometry of the original surface — and an isometric deformation of a developable surface into a state where regions lie flat against each other is a flat folding.

So the physics is solving, approximately, the constraint that this whole subject is built on: the sheet may not stretch. Given that constraint, the flat-folding conditions are not additional requirements the shell happens to satisfy; they are consequences of the constraint it is already under.

How nearly an isometry, in numbers #

“Very nearly an isometry” is the load-bearing phrase in the argument above, and it has a size, which is worth computing because the size is what makes the correspondence look exact rather than approximate.

For a thin shell the energy of bending scales with the cube of the thickness and the energy of stretching only with the thickness itself. At comparable deformations the ratio of the two is about $(t/R)^{2}$ — the thickness over the characteristic length, squared.

A drink can is about a tenth of a millimetre thick and thirty-three millimetres in radius, so that ratio is around $10^{-5}$. Bending is a hundred thousand times cheaper than stretching, and a minimiser that had any stretching in it would be paying five orders of magnitude for the privilege.

So the buckled mode is an isometry not approximately but to about one part in a hundred thousand. Kawasaki’s two alternating sums at a vertex of a real crushed can differ by something like a thousandth of a degree — far below what a protractor, a photograph or a caliper resolves, and far below the irregularities of the can itself.

Which is why the correspondence looks exact #

That number explains something the essay would otherwise have to leave as a coincidence: why the pattern satisfies the conditions exactly rather than nearly.

The conditions are equalities. A deformation that is 99 per cent isometric would give a crease pattern whose angles miss Kawasaki by a per cent, and a per cent of 180° is nearly two degrees — visible, measurable, and enough to make “the buckling mode is a flat-foldable pattern” a loose analogy rather than a statement. At one part in $10^{5}$ it is a statement.

The thinness of the shell is therefore not incidental to the finding; it is the finding’s whole tolerance budget. A thick-walled cylinder buckles too, and its mode is not a flat folding, because at $t/R$ of a tenth the ratio is only a hundred and stretching is affordable.

It also says where the correspondence stops on the can itself. Shell theory puts a boundary layer of width about $\sqrt{Rt}$ at any edge or clamped end, and for these numbers that is under two millimetres. Inside that band the deformation genuinely stretches and the crease pattern genuinely does not satisfy the conditions; outside it, which is nearly all of the can, it does.

So the honest boundary of the claim is spatial as well as idealised. The diamond lattice is a flat-foldable crease pattern everywhere except within a couple of millimetres of the rim, and the reason is a length scale rather than a hedge.

Two failures, one geometry #

The 1951 paper is about failure and that framing shaped what was recorded, which is worth a paragraph because it explains the twenty-year silence.

An aeronautical engineer wants to know the load at which the diamond mode appears, because that load is the design limit of the structure. The geometry of the mode is a means to that number: it is derived, used, and not dwelt on. Nothing in that literature asks whether the pattern folds flat, because flat-foldability is not a property anybody designing a fuselage has a use for.

A folder wants the opposite. The load is irrelevant; the geometry and its assignment are the entire content.

So the two communities extracted different things from the same object, and each left behind exactly what the other needed. That is a sharper version of the general point about search failures: the paper was not merely in the wrong venue, it emphasised the wrong half of its own result.

The assignment is the hard half #

This is where the essay’s real claim is, and it is stronger than the usual “nature got there first”.

A crease pattern without a mountain-and-valley assignment is not a folded state. Assigning letters to creases so that the result folds is the difficult part, the number of valid assignments is a vanishing fraction of all assignments, and the assignment most people would draw for a standard base cannot fold flat.

The buckled shell produces the assignment too. Which creases go outward and which inward is determined by the deformation, and what comes out satisfies Maekawa’s three-to-one split at every vertex.

What a designer would have had to do #

To see the size of what the shell handed over, it is worth costing the alternative.

A designer wanting a deployable surface with the Miura’s properties has to choose a repeating vertex, choose the sector angles, choose the assignment, and then verify that the result folds flat, has one degree of freedom, and is rigid-foldable. The first three are a search over a large space; the fourth is a set of conditions that most candidates fail.

Doing that in 1970 without a computer, from scratch, is a substantial piece of work. Reading the answer off a buckled shell is an afternoon. That is the sense in which the pattern was a gift rather than a discovery.

What the twenty years contained #

Yoshimura’s paper is of 1951 and is in aeronautics. The pattern enters the folding literature as a folding object much later, and Miura’s own work on the related fold dates from about 1970.

The mechanism is the one Beloch’s paper illustrates: a correct, published, findable result addressed to readers who were not the ones who eventually needed it most. A structural engineer studying shell collapse and a folder studying tessellations had no shared literature, no shared vocabulary, and no reason to search each other’s.

The difference here is that the two communities were both, in a sense, right about what the object was for. To aeronautics it is a failure mode — a thing to be prevented. To folding it is a mechanism — a thing to be used. The same geometry, with opposite sign.

The Miura fold’s date #

The related and more famous case deserves the same treatment.

The Miura fold is popularly dated to 1995, the year a satellite deployed a solar array folded that way. Koryo Miura described it in 1970, in work on pseudo-cylindrical concave polyhedral shells — which is to say, in the same aeronautical tradition and about the same class of problem.

So the twenty-five-year gap is not a gap in knowledge. The pattern was published and understood; what took twenty-five years was somebody having a solar array to fold.

The pattern has properties nobody put in #

What makes both patterns valuable is a set of properties that fall out rather than being designed.

The Miura has one degree of freedom: the whole sheet opens and closes with a single parameter, which is what makes it deployable by a single actuator. It has a negative Poisson’s ratio, so it contracts in both directions at once. It is rigid-foldable, so it works in panels rather than paper.

None of those was a design goal in 1970. They are consequences of the geometry, and the geometry came from a minimisation.

Crumpling is the same argument, weaker #

There is a version of this that goes too far and it is worth marking the boundary.

Crumple a sheet of paper into a ball and open it out, and the result is a crease pattern which — at the vertices where creases actually meet — satisfies the local conditions, for the same reason: it just folded, so a folding exists. That is an argument this site has already made and it is sound.

What crumpling does not give is structure. The pattern is irregular, has no repeating unit, has no useful degrees of freedom, and cannot be described more compactly than by listing its creases. It is flat-foldable and it is not a design.

The Yoshimura pattern is interesting precisely because it is both. The minimisation that produced it was translation-invariant, so the answer repeats, and a repeating answer is a pattern in the sense that matters — something that can be described, scaled, parameterised and manufactured.

The general shape of “found” #

It is worth being careful about what “found rather than designed” means, because it can be read as mysticism and is not.

A physical minimisation and a design process are both searches over the same space of configurations, with different objective functions. The buckling shell minimises elastic energy; a designer minimises something like number of creases or fits my requirements. When the two objectives happen to agree — as they do when both are dominated by the no-stretching constraint — the searches land in the same place.

So the physics is not smarter. It is optimising a related quantity, exhaustively, in constant time, over a space a person searches slowly.

The direction of the transfer, since 1970 #

Once the exchange was noticed it did not stop, and it has run mostly the other way since.

Folding now supplies patterns to engineering rather than taking them from it. Deployable arrays, airbag folding, stents, and packaging all use patterns chosen by folders or by design algorithms and handed to manufacturers, and the constraint that makes them work — one degree of freedom, rigid-foldable, flat-packing — is a folding constraint rather than a mechanical one.

So the 1951 borrowing is not the normal relationship between the two fields; it is the first exchange in a relationship that later reversed. Reading it as evidence that folding is derivative of engineering gets the sign wrong for everything after it.

What the figure checks #

The assertions are worth naming because the essay’s claim is unusual for a history essay: it is fully checkable.

The pattern is built from its construction and run through the vertex filter, which tests developability, Kawasaki, Maekawa and the big-little-big lemma at every interior vertex. The generator then separately asserts that every interior vertex has the same degree — establishing that it is a repeating unit and not a lucky drawing — and that the three-to-one split holds everywhere.

If the buckled geometry stopped having those properties the build would stop. That is a much stronger footing than any dated claim in this field can stand on.

What this does to the word invention #

Worth being precise, because the essay could be read as deflating two reputations and it is not.

Yoshimura did not invent the pattern; he identified and analysed a mode. Miura did not invent his either in the sense of choosing it from nothing — he was working in the same tradition, on the same class of problem, and arrived at a pattern with the properties that class of problem rewards. Both did substantial work and neither was a matter of noticing something obvious.

What they invented is the description: the parameterisation, the statement of which geometric quantities are free, and the account of what the resulting surface does. A physical process produces one instance. Turning an instance into a family that can be scaled, tuned and manufactured is the intellectual content, and it is not the sort of thing a buckling load does.

A note on what “lowest energy” is doing #

One clarification, since the phrase does a lot of work above.

The diamond mode is the lowest-energy failure among the modes considered in a linear buckling analysis, which is a statement about a particular idealised problem — a perfect cylinder, a perfect load, small deformations. Real shells have imperfections, and imperfection sensitivity is the defining feature of cylindrical buckling: the actual collapse load of a real can is a fraction of the theoretical one, and which mode appears depends on where the dents are.

So the pattern is the answer to a clean question that a real can approximates. That is enough for the argument here, which is about the geometry rather than about the load, but it is not the same as saying every crushed can produces it.

The idealisation, named #

The pattern as drawn is an idealisation of what a can does, and the gap is honest and large.

A real crushed cylinder has creases of finite radius, regions of genuine stretching near the ends, irregularities from imperfections in the shell, and a lattice that is often not quite closed. The Yoshimura pattern is the mode — the idealised, infinitely repeating, isometric deformation — and a real collapse approximates it.

So the claim is not that a drink can is a flat-foldable crease pattern. It is that the buckling mode is, and that a can approximates the mode well enough for the resemblance to be unmistakable.

Why the record entry runs backwards here #

This is one of the entries in the site’s record whose popular date is later than its evidence, and the reason is worth separating from the rest of the table.

Most claims in this field are dated too early because age is a credential and each retelling rounds down. These two run the other way, and the mechanism is different: the popular date attaches to the moment the object became visible to the public, which for the Miura fold is a satellite and for the diamond pattern is its arrival in the folding literature. Nobody is inflating anything. The date being quoted is a real date of a real event; it is simply not the date of the result.

That makes these entries useful as controls. A table in which every error had the same cause would be a table describing its compiler, and these two show the record’s overruns are produced by at least two independent mechanisms pointing in opposite directions.

Where this goes next #

The same vertex, found four times is the next rung: the observation that this is not an isolated case, and that a small number of vertex configurations keep being arrived at by unrelated routes.

The surprising connection: this essay’s finding inverts the usual relationship between craft and theory in the subject. Everywhere else, somebody folds something and the mathematics catches up. Here nothing folded anything — a load was applied and the paper found the answer — and both the craft and the theory arrived afterwards, separately, at something a shell had already solved.